PHAVEOS: The PHenology And Vegetation EO Service
T.H.G.Lankestera, J.Dashb, F.Baretc, S.R.Hubbarda
a
Astrium GEO Services., Farnborough, UK – (thomas.lankester, steven.hubbard)@infoterra-global.com
b
University of Southampton, Southampton, UK - j.dash@soton.ac.uk
c
INRA, Avignon, France - baret@avignon.inra.fr
Abstract - Monitoring vegetation biophysical variables is
key to understand the growth and functioning of vegetation
and in turn, provide key inputs in the modelling of bio
geochemical cycles. The PHenology And Vegetation EO
Service (PHAVEOS) realises the full spectral, spatial and
temporal resolution of the MERIS instrument by generating
of a range of biophysical variables on a daily basis. Level1b
data undergoes geometric, radiometric and atmospheric
correction before application of biophysical algorithms.
The resulting Level 2 data is then resampled and
composited to a 250m spatial grid to generate daily Level 3
output products. Modelling is then applied to create a time
series of Level 4 biophysical maps which retain daily
temporal resolution but with full spatial coverage.
PHAVEOS is laying the foundations for services that utilise
the even higher spatial and temporal resolutions available
from the upcoming Sentinel 2 and 3 series of EO platforms.
Keywords: vegetation, time-series, phenology, compositing,
modelling, monitoring, super-spectral
1. INTRODUCTION
The Medium Resolution Imaging Spectrometer (MERIS)
entered service with the launch of the European Space Agency
(ESA) ENVISAT platform in 2002. Following the arrival of the
Artemis data relay satellite in 2004, MERIS Full (300m)
Resolution data began to be systematically acquired providing
coverage every 2-3 days across 15 narrow wavebands covering
the visible to Near Infra-Red (NIR) part of the EM spectrum.
This unique combination of spectral, spatial and temporal
resolutions can support regular monitoring of a range of
vegetation indices and parameters that are key to understand the
growth and functioning of vegetation and, in turn, provide
inputs in the modelling of bio-geochemical cycles.
To meet this potential, ESA provided seed funding for the
development of the PHenology And Vegetation Earth
Observation Service (PHAVEOS).
Development was
undertaken by Astrium GEO Services in partnership with
University of Southampton and INRA.
1.1 Instrument capability
MERIS has the ability to monitor the ‘red edge’ (Baret, et al.
1992, Curran and Steele, 2005) where the absorption of light by
vegetation declines rapidly. As the total canopy chlorophyll
content increases, the absolute location of the red edge shifts to
longer wavelengths in the spectrum. The MERIS Terrestrial
Chlorophyll Index (MTCI) was developed (Dash and Curran,
2004) to capitalise on this phenomenon, so that MTCI remains
responsive to productivity even for high biomass (EspañaBoquera et al. 2006, Rossini et al. 2007) where more traditional
vegetation indices become insensitive. MTCI has also been
shown to have a superior correlation with Gross Primary
Productivity (GPP) than traditional measures (Wu et al. 2009,
Harris and Dash, 2010). MERIS can also be used as the basis
for the derivation of Essential Climate Variables (Anonymous,
2006) such as Leaf Area Index (LAI) and fraction of Absorbed
Photosynthetically Active Radiation (fAPAR) (Bacour, et al.
2006, Baret et al. 2006).
2. METHODS
In the first processing stage, standard Level 1b data are
geometrically, radiometrically and atmospherically corrected
before a range of (Level 2) biophysical parameters and indices
are derived. In the second stage, the biophysical data are
composited and mapped to one or more geographic grids using
an area weighted resampling algorithm (McGlynn, 2003).
Biophysical (Level 3) maps are produced for: LAI (Leaf Area
Index), MTCI (MERIS Terrestrial Chlorophyll Index), NDVI
(Normalised Difference Vegetation Index), fAPAR (fraction of
Absorbed Photosynthetically Active Radiation) and fCover
(fractional green cover).
The continuous spatial and temporal coverage required for timeseries of vegetation maps is traditionally achieved in two
discrete steps. Continuous spatial coverage is attained by
compositing multiple data acquisitions over time periods
ranging form a week to a month. Temporal interpolation and
smoothing is then applied to provide a continuous function of
how the mapped vegetation parameter or index varies with time.
The disadvantage of this traditional approach is that the
temporal resolution is constrained to the compositing period
chosen rather than to the fundamental frequency of data
acquisition. In some cases (e.g. melting snow cover) the use of
an extended compositing period can even result in a systematic
temporal bias.
To avoid temporal bias and maximise the temporal resolution,
PHAVEOS generates Level 3 composites on a daily basis.
These products are then used as the basis for a two step
modelling exercise. A spine interpolation of values is first
applied to give continuous set of daily values for each map grid
cell. A smoothing algorithm is then applied to generate the final
(Level 4) time-series of biophysical maps with continuous
250m, daily coverage.
The smoothing algorithm employed should be parameter
dependent.
A locally weighted scatter-plot smoothing
(MATLAB rloess regression) is applicable for the generation of
the MTCI Level 4 products however for products such as
NDVI, which exhibit a negative noise bias (Dash et al., 2008), a
maximum value polynomial smoothing is more appropriate.
As well as generating daily interpolated and smoothed
biophysical variable values per map grid cell, the Level 4
processing also generate three quality measures: time (in days)
to the closest observation derived value; value difference
between the current value and closes observation derived value
and finally, the difference between the current interpolated and
smoothed values. The two interpolation quality measures
reflect the availability of observations whereas the degree of
smoothing provides a measure of noise.
3. RESULTS AND APPLICATIONS
Sample Level 3 and Level 4 PHAVEOS products are illustrated
in Figures 1, 2 and 3.
Figure 3. 2009 interpolated (blue) and smoothed (red) MTCI
values for the oak woodland grid map cell outlined in Figure 2.
Figure 1. Level 3 (composite) map of fractional green cover
across Ireland for 14th January 2006.
The 250m spatial grid used for the PHAVEOS vegetation maps
is best appreciated in Figure 2.
Service development has been carried out with a number of user
partners in the UK and Ireland leading to the identification of a
range of potential applications.
PHAVEOS has already
demonstrated its worth for public education and engagement by
providing dynamic maps showing the dramatic difference
between the Spring greening events of 2009 and 2010 across the
British Isles (Nature’s Calendar, 2010). For forestry, canopy
and woodland health mapping, forest damage identification and
monitoring, and determination of growing season length are all
areas for further investigation. In nature conservation, the
service could assist with landscape-scale ecosystem assessment
and habitat surveillance, more specifically evaluating how
phenological information may help with distinguishing specific
habitat types otherwise difficult to distinguish from
conventional optical EO approaches. Work is also on-going to
adapt, from MODIS to MERIS, the approach of van der Wal
(2010) for the use of medium resolution (250 m pixel) imagery
to detect phenological changes occurring on the intertidal areas.
For all of these applications there is a basic need to characterise
both data availability and the ability to detect natural changes
and distinguish them from the effects of instrument,
atmospheric and processing induced noise. This requirements
applies to each of the vegetation parameters and indices
produced by PHAVEOS.
4. CHARACTERISATION AND VALIDATION
The availability of PHAVEOS products will be characterised
spatially and temporally by generating maps of the percentage
of data availability over a 6 year time span from 2005-2010.
An example of such map is provide in Figure 4, which
illustrates that data coverage in the northwest is relatively low
(less than 10%) compared to the southern part of the British
Isles. For areas having good data coverage, the reliability of the
products will be provided for key phenological stages such as
during spring and autumn. This will be done using a temporal
moving window to identify temporal gaps during these key
seasons.
Figure 2. fCover map for 19th March 2009, superimposed over
orthorectified aerial photography of Alice Holt Forest, UK
Derived from MERIS data acquired in early Spring 2009, prior
to bud-burst, the forest area in the centre of Figure 2 exhibits a
lower fractional green cover than the golf course located to the
north (top centre of Figure 2).
To illustrate phenology dynamics the modelled (Level 4) results
for a single map cell can be plotted through time (see Figure 3).
To assist with the understanding and interpretation of the high
temporal resolution Level 4 time series, the time series will be
characterised by two measures of noise:
(i) the change in observation derived biophysical
variables per day;
(ii) the smoothed value / difference between the smoothed
and interpolation values.
The first measure contains a degree of natural signal variation
whereas the second measure attempts to remove and normalises
for the signal, yielding a measure of signal to noise ratio (SNR).
In parallel with the basic characterisation of PHAVEOS time
series, work will be carried out in to relation to ground based
measurements to provide a better understanding of land surface
processes. Broadly three types of data can be used to validate
outputs from PHAVEOS:
(i) ground observations;
(ii) high spatial resolution satellite data;
(iii) digital camera images (from ‘phenocam’) obtained at
very high (30 minute) temporal resolution.
Where possible, biophysical products derived from PHAVEOS
will be compared with ground measurements obtained at
different time of the growing season. Then the standard error of
prediction (variation of PHAVEOS output with respect to the
ground data over a week) of the biophysical products will be
determined as a measure of accuracy of the product. For
woodland sites at Alice Holt Forest and Wytham Wood in the
UK, one year of phenocam data will be used to derive annual
time series of reflectance in red, green and blue. These time
series will be then compared with the PHAVEOS derived daily
time series of reflectance in red, green and blue bands.
Figure4. Illustrative data availability for 2008.
5. CONCLUSIONS
By making spatially and temporally continuous time series of a
number of vegetation variable available, PHAVEOS is enabling
the investigation of operational utilisation of super-spectral EO
data. This on-going investigation to characterise and validate
these products will lead to a targeted market development with
applications ranging from intertidal monitoring, land use change
detection, inventory of forest health and productivity, to support
for ecosystem / biodiversity assessment.
With 300m resolution MERIS data available from July 2004,
PHAVEOS is developing a multi-year set of time series that
will act as a baseline for continued service development based
on the Ocean and Land Colour Instrument (OCLI) due to be
deployed with the launch of the first upcoming Sentinel 3
platform in 2013.
The techniques and applications developed with MERIS data
will also form the basis for an expansion of PHAVEOS to new
applications that utilise the 20m resolution of the Sentinel-2
Multi-Spectral Imager (MSI).
REFERENCES
Anonymous, 2006. GCOS-107 (WMO/TD No. 1338)
“Supplemental details to the satellite based component of the
Implementation Plan for the Global Observing System for
Climate in Support of the UNFCCC" Geneve , Switzerland,
http://www.wmo.int/pages/prog/gcos/Publications/gcos-107.pdf
C. Bacour, F. Baret, et al., “Neural network estimation of LAI,
fAPAR, fCover and LAIxCab, from top of canopy MERIS
reflectance data: principles and validation”, Remote Sensing of
Environment 105, p.p. 313-325, 2006
F. Baret, S. Jacquemoud, G. Guyot, C. Leprieur, "Modeled
analysis of the biophysical nature of spectral shifts and
comparison with information content of broad bands. Remote
Sensing of the Environment, 41, p.p. 133-142., 1992
Baret, F. Pavageau, K., Beal, D., Weiss, M., Berthelot, B. and
Regner, P., 2006. “Algorithm Theoretical Basis Document for
MERIS Top of Atmosphere Land Products (TOA_VEG)
Version 3” Report of ESA contract AO/1-4233/02/I-LG,
http://www.brockmannconsult.de/beam/software/repositories/4.6/beam-meris-veg2.0.1/docs/MERIS_ATBD_TOA_VEG_03_06.pdf
Baret, F., Weiss, M., Garrigues, S., 2008. In WGCV-29,
Session 4: Land Product Validation Special Session, INRA /
CNES, Avignon, France. “OLIVE On Line Interactive
Validation Exercise”
http://www.ceos.org/images/wgcv/wgcv29/OLIVEPARaMETER_Baret.pdf
Baret, F., Billard, G., Marloie1, O., Hanocq1, J.F., de Solan, B.,
A. Labouret, A., 2010. CEOS LPV Phenology focus group,
“PAR@METER: a wireless system for fAPAR and LAI
continuous monitoring.”
http://lpvs.gsfc.nasa.gov/PDF/PAR@METER.pdf
P. J. Curran and C. M. Steele, “MERIS: The re-branding of an
ocean sensor”, International Journal of Remote Sensing, 26, p.p.
1781-1798, 2005
J. Dash and P. J. Curran,. “The MERIS terrestrial chlorophyll
index”. International Journal of Remote Sensing, 25, p.p. 54035413, 2004
J. Dash, J. Rumsey, T.H.G. Lankester, S.R. Hubbard and P.J.
Curran, “A comparison of two phenological variables derived
from VEGETATION NDVI and MERIS MTCI: A feasibility
study in the context of Global Monitoring for Environment and
Security (GMES)”. 37th COSPAR Scientific Assembly.
Montréal, Canada., p.p.646, July 2008
L.M. Espana-Boquera, R. Cardenas-Navarro, L. Lopez-Perez,
V. Castellanos-Morales and LOBIT, P. Lobit, “Estimating the
nitrogen concentration of strawberry plants from its spectral
response”, Communications in Soil Science and Plant Analysis,
37, p.p. 2447-2459, 2006
A. Harris and J. Dash, “An evaluation of the MERIS Terrestrial
Chlorophyll Index for direct estimation of carbon fluxes”,
Remote Sensing of Environment, 114, p.p. 1856-1862, 2010
McGlynn, T.A. 2003. In AAS 201st Meeting, Session. 91.05 “A
fast flux-conserving resampling algorithm”, Seattle, WA, USA,
http://aas.org/archives/BAAS/v34n4/aas201/895.htm?q=publica
tions/baas/v34n4/aas201/895.htm
Nature’s Calendar “Spring from outer space – June update”
http://www.naturescalendar.org.uk/findings/spring_from_outer_
space_june_update.htm
D. van der Wal, A. Wielemaker-van den Dool and P.M.J.
Herman,.” Spatial synchrony in intertidal benthic algal biomass
in temperate coastal and estuarine ecosystems”. Ecosystems, 13
p.p. 338-351, 2010
C.Y.Wu, Z.Niu, Q.Tang, W.J.Huang, B.Rivard and J.L.Feng,
“Remote estimation of gross primary production in wheat using
chlorophyll-related vegetation indices”, Agricultural and Forest
Meteorology, 149, p.p. 1015-1021, 2009
ACKNOWLGEMENTS
The authors would like to acknowledge the support of the ESA
EODM programme which funded the PHAVEOS development,
supplied data and generously donated access to the Grid
Processing On-Demand (G-POD) data processing infrastructure.
In particular, we would like to acknowledge the advice,
guidance and encouragement provided by Dr B.Koetz, the ESA
project officer.